Interpretable and Scalable Similarity Metrics for DNA-Encoded Library Design Using Generative Topographic Mapping

Area of Science:

• Medicinal Chemistry
• Computational Chemistry
• Cheminformatics

Background:

• DNA-encoded libraries (DELs) are expanding, necessitating efficient selection criteria.
• Selecting optimal DELs requires balancing similarity to active compounds and intra-library diversity.
• Existing methods for DEL selection lack scalability and interpretability.

Purpose of the Study:

• To evaluate Generative Topographic Mapping (GTM) as a scalable and interpretable method for selecting DNA-encoded libraries (DELs).
• To compare GTM-derived "stand-alone" (SA) metrics with conventional compound pair-matching (CP) metrics for DEL selection.
• To assess GTM's ability to identify DELs that are both reference-proximal and chemically diverse.

Main Methods:

• Comparative analysis of Morgan count fingerprint-based CP metrics and GTM-derived SA metrics.
• Utilized 100 diverse DEL subsets and a ChEMBL reference set for cyclin-dependent kinase 2 (CDK2) screening.
• Calculated Spearman rank correlations and enrichment factors (EF5%) to assess metric performance.

Main Results:

• GTM-based SA metrics demonstrated robust approximations of gold standard CP metrics, with Spearman correlations ranging from 0.6-0.7.
• GTM successfully identified DELs that best span the reference space, achieving EF5% values of 4-12.
• GTM consistently selected top-performing libraries, identifying 2 out of the top 3 gold standard libraries within the top 5% ranked by GTM.
• Two-dimensional GTM landscapes provided visual assessment of intra- and inter-library diversity.

Conclusions:

• GTM serves as an efficient tool for chemical library similarity assessment and target-focused DEL selection.
• GTM offers a scalable and interpretable alternative to traditional methods for designing and screening DELs.
• The visual outputs of GTM support rapid evaluation of diverse DEL designs for drug discovery.